EigenGauntlet

EigenGauntlet is an EigenScript stress application suite for constrained hardware. It is intentionally built as a set of small real workloads rather than synthetic microbenchmarks.

The goal is to expose EigenScript language, runtime, standard library, and tooling gaps without hiding them behind downstream workarounds. When a workload needs a primitive EigenScript should own, the issue goes in GAPS.md and gets fixed at the root.

Requirements

• EigenScript v0.26.0 (built from source; CI tests against this version)
• No external runtime dependencies for the default smoke suite

Use the local development binary directly:

../EigenScript/src/eigenscript gauntlet.eigs --suite smoke

Or use an installed binary:

eigenscript gauntlet.eigs --suite smoke

Workloads

• tensor - batched neural-layer math, matrix/vector broadcasting, softmax
• dataflow - JSON, strings, dictionaries, file I/O, EigenStore
• swarm - real-time style entity simulation and spatial queries
• swarm-batch - batched spatial queries via nearest_in_range_all
• concurrent - threads, channels, non-blocking receive
• parser - try_parse, eval, generated source, functions, match, f-strings
• memory - list/dict/tensor churn, closures, arena mark/reset
• observer - interrogatives, predicates, report states, unobserved blocks
• io - stream tensor I/O, buffers, byte reads, tensor save/load
• realtime - frame-budget style fixed-step loop with monotonic timing
• module-scope - eval, load_file, import isolation, module closures

Hardware Discipline

This repo is designed for a 2007 dual-Pentium laptop. Defaults must stay small enough to run quickly on that hardware. Larger modes can exist, but they should be opt-in and named clearly.

Commands

# Run all smoke workloads
./tests/run_smoke.sh

# Run each workload separately with timings
./tests/run_each.sh

# Run one workload
../EigenScript/src/eigenscript gauntlet.eigs --suite tensor

# Run a bigger local pass
../EigenScript/src/eigenscript gauntlet.eigs --suite smoke --size 3